git: b7e05a72e644 - 2023Q1 - graphics/openexr: Fix ARMv7 build
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Tue, 21 Mar 2023 19:23:54 UTC
The branch 2023Q1 has been updated by mandree:
URL: https://cgit.FreeBSD.org/ports/commit/?id=b7e05a72e644249c32c2e723d015b7af444c8690
commit b7e05a72e644249c32c2e723d015b7af444c8690
Author: Matthias Andree <mandree@FreeBSD.org>
AuthorDate: 2023-03-21 16:42:19 +0000
Commit: Matthias Andree <mandree@FreeBSD.org>
CommitDate: 2023-03-21 19:22:30 +0000
graphics/openexr: Fix ARMv7 build
OpenEXR 3.1.6 introduced several NEON-based optimizations
that implied Aarch64. Add patched, either picked from
upstream, or written by mandree@, to enable those
NEON features that also require Aarch64 only there.
PR-1366 is cherry-picked from upstream, and patch-lib/patch-test files
are my work but build upon said PR.
Also cherry-pick PR1354 that adds a missing check for AVX,
which is why I am bumping PORTREVISION because it might change
code (I have not checked).
https://github.com/AcademySoftwareFoundation/openexr/issues/1365
PR: 270348
Reported by: fuz@ (Robert Clausecker)
(cherry picked from commit 54d6860aac464b690f37007cfaef52c5a0b427b0)
---
graphics/openexr/Makefile | 4 +-
...PR1354-3f97750d1ec203e7d7eb8d5f30f3d5e7e68ad720 | 25 ++++++
...PR1366-a41a736d64e3d93baffef1042d4a3d1aaf74f1c9 | 98 ++++++++++++++++++++++
.../files/patch-lib_OpenEXR_ImfDwaCompressor.cpp | 11 +++
.../files/patch-lib_OpenEXR_ImfDwaCompressorSimd.h | 29 +++++++
...atch-test_OpenEXRTest_testDwaCompressorSimd.cpp | 38 +++++++++
6 files changed, 203 insertions(+), 2 deletions(-)
diff --git a/graphics/openexr/Makefile b/graphics/openexr/Makefile
index 87011420e1b1..26c735c4849f 100644
--- a/graphics/openexr/Makefile
+++ b/graphics/openexr/Makefile
@@ -1,6 +1,6 @@
PORTNAME= openexr
PORTVERSION= 3.1.6
-PORTREVISION= 0
+PORTREVISION= 1
CATEGORIES= graphics devel
MAINTAINER= mandree@FreeBSD.org
@@ -23,7 +23,7 @@ CPPFLAGS+= -I.
# must be linked with -l{thr|pthread} explicitly
LDFLAGS+= -lpthread
-PATCH_STRIP= -p0
+PATCH_STRIP= -p1
_MAJORVER= 3_1
_VER= 30
diff --git a/graphics/openexr/files/patch-PR1354-3f97750d1ec203e7d7eb8d5f30f3d5e7e68ad720 b/graphics/openexr/files/patch-PR1354-3f97750d1ec203e7d7eb8d5f30f3d5e7e68ad720
new file mode 100644
index 000000000000..edd1bb3ecc08
--- /dev/null
+++ b/graphics/openexr/files/patch-PR1354-3f97750d1ec203e7d7eb8d5f30f3d5e7e68ad720
@@ -0,0 +1,25 @@
+From 3f97750d1ec203e7d7eb8d5f30f3d5e7e68ad720 Mon Sep 17 00:00:00 2001
+From: Cary Phillips <cary@ilm.com>
+Date: Thu, 16 Mar 2023 07:23:18 -0700
+Subject: [PATCH] Test for AVC in unpack.c (#1354)
+
+Addresses #1353
+
+Signed-off-by: Cary Phillips <cary@ilm.com>
+---
+ src/lib/OpenEXRCore/unpack.c | 2 +-
+ 1 file changed, 1 insertion(+), 1 deletion(-)
+
+diff --git a/src/lib/OpenEXRCore/unpack.c b/src/lib/OpenEXRCore/unpack.c
+index 606d7c3eb..ac4e259f3 100644
+--- a/src/lib/OpenEXRCore/unpack.c
++++ b/src/lib/OpenEXRCore/unpack.c
+@@ -37,7 +37,7 @@ half_to_float8 (float* out, const uint16_t* src)
+ }
+ #endif
+
+-#if (defined(__x86_64__) || defined(_M_X64)) && \
++#if (defined(__x86_64__) || defined(_M_X64)) && defined(__AVX__) && \
+ (defined(__F16C__) || defined(__GNUC__) || defined(__clang__))
+
+ # if defined(__F16C__)
diff --git a/graphics/openexr/files/patch-PR1366-a41a736d64e3d93baffef1042d4a3d1aaf74f1c9 b/graphics/openexr/files/patch-PR1366-a41a736d64e3d93baffef1042d4a3d1aaf74f1c9
new file mode 100644
index 000000000000..054dc785eec5
--- /dev/null
+++ b/graphics/openexr/files/patch-PR1366-a41a736d64e3d93baffef1042d4a3d1aaf74f1c9
@@ -0,0 +1,98 @@
+From a41a736d64e3d93baffef1042d4a3d1aaf74f1c9 Mon Sep 17 00:00:00 2001
+From: Aras Pranckevicius <aras@nesnausk.org>
+Date: Mon, 20 Mar 2023 19:31:42 +0200
+Subject: [PATCH] Fix ARMv7 build by making recent ZIP NEON optimizations be
+ ARMv8 (aarch64) only
+
+Should fix #1365. Recent PR (#1348) added NEON accelerated code paths
+for ZIP filtering. But that code uses several instructions that are
+ARMv8 (aarch64) only, and thus fail building on 32-bit ARM (armv7)
+platforms. Make these optimizations only kick in when building
+for 64-bit ARM platforms.
+
+Signed-off-by: Aras Pranckevicius <aras@nesnausk.org>
+---
+ src/lib/OpenEXR/ImfSimd.h | 4 ++++
+ src/lib/OpenEXR/ImfZip.cpp | 6 +++---
+ src/lib/OpenEXRCore/internal_zip.c | 8 ++++----
+ 3 files changed, 11 insertions(+), 7 deletions(-)
+
+diff --git a/src/lib/OpenEXR/ImfSimd.h b/src/lib/OpenEXR/ImfSimd.h
+index 3053a5d4e..d1f064525 100644
+--- a/src/lib/OpenEXR/ImfSimd.h
++++ b/src/lib/OpenEXR/ImfSimd.h
+@@ -46,6 +46,10 @@
+ # define IMF_HAVE_NEON
+ #endif
+
++#if defined(__aarch64__)
++# define IMF_HAVE_NEON_AARCH64 1
++#endif
++
+ extern "C" {
+ #ifdef IMF_HAVE_SSE2
+ # include <emmintrin.h>
+diff --git a/src/lib/OpenEXR/ImfZip.cpp b/src/lib/OpenEXR/ImfZip.cpp
+index 0e2b031d8..8dd53bea9 100644
+--- a/src/lib/OpenEXR/ImfZip.cpp
++++ b/src/lib/OpenEXR/ImfZip.cpp
+@@ -160,7 +160,7 @@ reconstruct_sse41 (char* buf, size_t outSize)
+
+ #endif
+
+-#ifdef IMF_HAVE_NEON
++#ifdef IMF_HAVE_NEON_AARCH64
+
+ void
+ reconstruct_neon (char* buf, size_t outSize)
+@@ -262,7 +262,7 @@ interleave_sse2 (const char* source, size_t outSize, char* out)
+
+ #endif
+
+-#ifdef IMF_HAVE_NEON
++#ifdef IMF_HAVE_NEON_AARCH64
+
+ void
+ interleave_neon (const char* source, size_t outSize, char* out)
+@@ -380,7 +380,7 @@ Zip::initializeFuncs ()
+ }
+ #endif
+
+-#ifdef IMF_HAVE_NEON
++#ifdef IMF_HAVE_NEON_AARCH64
+ reconstruct = reconstruct_neon;
+ interleave = interleave_neon;
+ #endif
+diff --git a/src/lib/OpenEXRCore/internal_zip.c b/src/lib/OpenEXRCore/internal_zip.c
+index e829e6d7c..1374cab0d 100644
+--- a/src/lib/OpenEXRCore/internal_zip.c
++++ b/src/lib/OpenEXRCore/internal_zip.c
+@@ -24,8 +24,8 @@
+ # define IMF_HAVE_SSE4_1 1
+ # include <smmintrin.h>
+ #endif
+-#if defined(__ARM_NEON)
+-# define IMF_HAVE_NEON 1
++#if defined(__aarch64__)
++# define IMF_HAVE_NEON_AARCH64 1
+ # include <arm_neon.h>
+ #endif
+
+@@ -78,7 +78,7 @@ reconstruct (uint8_t* buf, uint64_t outSize)
+ prev = d;
+ }
+ }
+-#elif defined(IMF_HAVE_NEON)
++#elif defined(IMF_HAVE_NEON_AARCH64)
+ static void
+ reconstruct (uint8_t* buf, uint64_t outSize)
+ {
+@@ -174,7 +174,7 @@ interleave (uint8_t* out, const uint8_t* source, uint64_t outSize)
+ *(sOut++) = (i % 2 == 0) ? *(t1++) : *(t2++);
+ }
+
+-#elif defined(IMF_HAVE_NEON)
++#elif defined(IMF_HAVE_NEON_AARCH64)
+ static void
+ interleave (uint8_t* out, const uint8_t* source, uint64_t outSize)
+ {
diff --git a/graphics/openexr/files/patch-lib_OpenEXR_ImfDwaCompressor.cpp b/graphics/openexr/files/patch-lib_OpenEXR_ImfDwaCompressor.cpp
new file mode 100644
index 000000000000..5b908fdad69e
--- /dev/null
+++ b/graphics/openexr/files/patch-lib_OpenEXR_ImfDwaCompressor.cpp
@@ -0,0 +1,11 @@
+--- ./src/lib/OpenEXR/ImfDwaCompressor.cpp.orig 2023-03-05 20:23:47 UTC
++++ ./src/lib/OpenEXR/ImfDwaCompressor.cpp
+@@ -2888,7 +2888,7 @@ DwaCompressor::initializeFuncs()
+ fromHalfZigZag = fromHalfZigZag_f16c;
+ }
+
+-#ifdef IMF_HAVE_NEON
++#ifdef IMF_HAVE_NEON_AARCH64
+ {
+ convertFloatToHalf64 = convertFloatToHalf64_neon;
+ fromHalfZigZag = fromHalfZigZag_neon;
diff --git a/graphics/openexr/files/patch-lib_OpenEXR_ImfDwaCompressorSimd.h b/graphics/openexr/files/patch-lib_OpenEXR_ImfDwaCompressorSimd.h
new file mode 100644
index 000000000000..e9eff7e1a909
--- /dev/null
+++ b/graphics/openexr/files/patch-lib_OpenEXR_ImfDwaCompressorSimd.h
@@ -0,0 +1,29 @@
+--- ./src/lib/OpenEXR/ImfDwaCompressorSimd.h.orig 2023-03-05 20:23:47 UTC
++++ ./src/lib/OpenEXR/ImfDwaCompressorSimd.h
+@@ -395,7 +395,7 @@ convertFloatToHalf64_scalar (unsigned short* dst, floa
+ dst[i] = ((half) src[i]).bits ();
+ }
+
+-#ifdef IMF_HAVE_NEON
++#ifdef IMF_HAVE_NEON_AARCH64
+
+ void
+ convertFloatToHalf64_neon (unsigned short* dst, float* src)
+@@ -821,7 +821,7 @@ fromHalfZigZag_f16c (unsigned short* src, float* dst)
+ #endif /* defined IMF_HAVE_GCC_INLINEASM_X86_64 */
+ }
+
+-#ifdef IMF_HAVE_NEON
++#ifdef IMF_HAVE_NEON_AARCH64
+
+
+ void
+@@ -856,7 +856,7 @@ fromHalfZigZag_neon(unsigned short* __restrict__ src,
+ }
+ }
+
+-#endif // IMF_HAVE_NEON
++#endif // IMF_HAVE_NEON_AARCH64
+
+ //
+ // Inverse 8x8 DCT, only inverting the DC. This assumes that
diff --git a/graphics/openexr/files/patch-test_OpenEXRTest_testDwaCompressorSimd.cpp b/graphics/openexr/files/patch-test_OpenEXRTest_testDwaCompressorSimd.cpp
new file mode 100644
index 000000000000..4d9991f463d3
--- /dev/null
+++ b/graphics/openexr/files/patch-test_OpenEXRTest_testDwaCompressorSimd.cpp
@@ -0,0 +1,38 @@
+--- ./src/test/OpenEXRTest/testDwaCompressorSimd.cpp.orig 2023-03-05 20:23:47 UTC
++++ ./src/test/OpenEXRTest/testDwaCompressorSimd.cpp
+@@ -401,7 +401,7 @@ testFloatToHalf()
+ }
+ }
+
+-#ifdef IMF_HAVE_NEON
++#ifdef IMF_HAVE_NEON_AARCH64
+ {
+ cout << " convertFloatToHalf64_neon()" << endl;
+ for (int iter = 0; iter < numIter; ++iter)
+@@ -430,7 +430,7 @@ testFloatToHalf()
+ }
+ }
+ }
+- #endif // IMF_HAVE_NEON
++ #endif // IMF_HAVE_NEON_AARCH64
+ }
+
+ //
+@@ -516,7 +516,7 @@ testFromHalfZigZag()
+ } // iter
+ } // f16c
+
+-#ifdef IMF_HAVE_NEON
++#ifdef IMF_HAVE_NEON_AARCH64
+ {
+ const int numIter = 1000000;
+ Rand48 rand48 (0);
+@@ -553,7 +553,7 @@ testFromHalfZigZag()
+ } // iter
+ } // neon
+
+-#endif // IMF_HAVE_NEON
++#endif // IMF_HAVE_NEON_AARCH64
+ }
+
+